Answer the question
In order to leave comments, you need to log in
Is it possible to recognize the number of the car from the video?
There is an original video from the camera. This is a screenshot from the video .
Is there any way to recognize the number of the car?
Answer the question
In order to leave comments, you need to log in
If we are talking about what, like in a movie, to make a good photo out of a bad one, then no, you can’t.
Machine learning algorithms are able to “guess” what was probably meant and, for example, increase the resolution of a photo if they were previously “showed” similar areas of images. But in the problem with number, the approach is narrower.
There is additional data:
it is known that the number is there, what format it is, what possible characters it consists of, and there is not a single frame, but many - all the frames of the video where the same number got, from different angles.
Therefore, it is probably possible to train the model on examples of other numbers and, as a result, suggest some of the most probable variants of the desired number.
As in face recognition, first find the area with the number in the image and normalize it by “expanding” so that the number is exactly occupied by a rectangle that will be the entrance to the second network.
If the number is less than a pixel, then it is completely unrealistic.
There are algorithms that allow you to recognize a license plate from a moving video when the number falls on different pixels in different frames. However, the reliability is still very low.
Start by viewing the frame in the graphic editor - zoom in to a pixel size of a couple of millimeters. I have poor eyesight so I can't take it.
Perhaps a neural network, for example, can read frames from a video and identify geometrically similar objects with a license plate, and then cut out the numbers inside. You just need to write algorithms to determine the signs at different angles. Nothing complicated
Didn't find what you were looking for?
Ask your questionAsk a Question
731 491 924 answers to any question